ximations are accepted and a completely rigorous solution is disregarded in favour of a some
what simpler treatment. The leading idea then is to work with independent units which undergo
no other alterations or deformations than by transformation (including non-linear transforma
tion). Independent models (stereopairs) are the most familiar example. The choice of suitable
units must not only consider the computing technique, but should rather be based on the feasi
bility from the photogrammetric point of view. In case units are chosen whose approximations
go too far (whose internal residual errors cannot be tolerated) it is difficult to rectify the over
simplification, at least with direct solution procedures. Application of relaxation techniques
might again still be feasible, however.
At present the main concentration seems to be on adjustment procedures based on the
treatment of independent models (as a special case of independent sections). The approxima
tions are very small compared with the ideal single plates approach. A theoretical investiga -
tion into the effect of the approximation is, however, still missing. Mathematical formulations
of the model approach are given in [7 0], [74], [75], [83] - [86]. The use of models as basic ad
justment units has the advantage that its application is not restricted to comparator measure -
ments and also that models measured in any stereo-restitution instrument can go into the block-
adjustment directly. Finally, but certainly not the least practical advantage, is the fact that the
adjusted units are directly suited for subsequent plotting. Under the influence of this kind of
block-adjustment the conventional strip-triangulation and hence the first order instruments
seem to loose importance in favour of measurement of independent models in second order ins
truments supplied with coordinate registration equipment.
The obvious successes of the model approach of block-adjustment reduce the practical
importance of those block-adjustment procedures which work with polynomial corrections for
strips and which are labeled as distinctly approximate. This is especially true since in both
cases middle class computers are suited and the amount of computing required is apparently of
a similar order of magnitude. Hence the interest in less accurate and less general applicable
polynomial procedures will probably decrease in future.
*3. 5 Regarding the task of numerical solution of block-adjustment it is remarkable that all
theoretical approaches based on any form of independent units lead essentially to the same type
of computing problem. The number of units may be different but the common problem is to
connect them by taking into account all mutual interrelations. It can be shown that the various
approaches give, or can be reduced to, systems of equations of siriiilar structure. The non
zero terms of the equations to be solved can be arranged to form a band along the principal dia
gonal of the coefficient matrix. The width of the band is determined by the number of units
which are directly interconnected. In [86] the author has shown the structural identity of the
normal equations or partially reduced normal equations of some approaches. This property is
even independent of the actual choice of the unknowns.
All considerations about the principles of numerical solution of block-adjustment start
from the basic fact that the number of unknowns to be determined is usually rather large (or -
der of magnitude 10^ - 10^ ). The great number of solution techniques available can be classi
fied in a few groups :
a) The direct solution working with an elimination procedure according to the Gauss algorith-
me or other equivalent reduction techniques. It seems that the direct solution has received lit
tle attention, probably because of apparent storage requirements. Only the ITC procedure fol
lows at present this line. The existence of a direct solution which can favourably compete with
other solutions might influence the future development.
b) The iterative solution of a large number of equations seems in general to be the most attrac
tive procedure both for large and medium size computers. With iterative solutions the problem
of convergence arises. In blocks with little ground control the convergence is slow. The Ord
nance survey and others ( [56], [7 0] ) have collected experimental information about the rate of
convergence of iterative solutions for the block sizes normally encountered. The drawback of
it is that the results of experiments are only to be trusted within the range investigated but are
still not sufficiently informative about unusual cases. Miles [85] has referred to a theoretical